A FastAPI service for searching college articulation PDFs using Pinecone vector database. Returns raw search results designed to be used as a tool by Gemini or other LLMs.

• 📄 Automatic PDF text extraction and chunking
• 🔍 Semantic search using Pinecone with automatic embeddings (llama-text-embed-v2)
• 🚀 FastAPI REST endpoints
• 📊 Source citations and similarity scores for each chunk
• 🧩 Tool-ready: Returns structured data perfect for LLM integration

1. Install dependencies:

pip install -r requirements.txt

2. Set environment variables: Create a .env file with:

PINECONE_API_KEY=your_pinecone_api_key_here

3. Run the API:

python main.py

Or with uvicorn directly:

uvicorn main:app --reload --host 0.0.0.0 --port 8000

Health check endpoint.

Detailed health check with index statistics.

Ingest PDF(s) into Pinecone vector database.

Request body (optional):

{
  "pdf_path": "pdfs/AA_UF.pdf"  // Optional: specific PDF path. If omitted, ingests all PDFs in pdfs/ folder
}

Response:

{
  "message": "Successfully ingested 1 PDF(s) with 150 total chunks",
  "ingested_count": 150,
  "pdfs_processed": ["AA_UF"]
}

Search the college PDF vector database and return relevant chunks.

Request body:

{
  "query": "What are the articulation agreements for University of Florida?",
  "top_k": 5  // Optional: number of chunks to retrieve (default: 5)
}

Response:

{
  "query": "What are the articulation agreements for University of Florida?",
  "chunks": [
    {
      "text": "The University of Florida has articulation agreements...",
      "source": "AA_UF",
      "score": 0.92,
      "id": "AA_UF_0"
    },
    {
      "text": "Transfer credits from community colleges...",
      "source": "AA_UF",
      "score": 0.88,
      "id": "AA_UF_1"
    }
  ],
  "total_results": 2,
  "context": "[Source: AA_UF]\nThe University of Florida has articulation agreements...\n\n---\n\n[Source: AA_UF]\nTransfer credits from community colleges..."
}

Get statistics about the Pinecone index.

1. Ingest all PDFs:

curl -X POST http://localhost:8000/ingest

2. Search the vector database:

curl -X POST http://localhost:8000/search \
  -H "Content-Type: application/json" \
  -d '{"query": "What are the articulation agreements for Florida universities?", "top_k": 5}'

This API is designed to be used as a tool by Gemini. The response format provides:

• chunks: Array of relevant document chunks with metadata
• context: Pre-formatted combined context string ready for LLM consumption
• sources: Source PDF names for citation
• scores: Similarity scores for relevance ranking

Example Gemini function calling:

# Gemini can call this API and use the returned chunks/context
# to generate answers based on the retrieved documents

• Pinecone: Vector database with automatic embeddings using llama-text-embed-v2
• PyPDF2: PDF text extraction
• FastAPI: REST API framework

• The index is automatically created if it doesn't exist
• PDFs are chunked with 1000 character chunks and 200 character overlap
• Each chunk includes metadata with the source PDF name
• The API automatically handles embedding generation through Pinecone's integrated model
• Returns raw search results - no LLM integration (designed to be used as a tool)